Outcome prediction in patients with moderate and severe traumatic brain injury using machine learning models – a Big Data approach in modern healthcare analysis

This abstract has open access
Abstract Description
Abstract ID :
HAC1758
Submission Type
Authors (including presenting author) :
MAK Hoi Kwan Calvin(1), YUAN Yixuan(2), GUO Xiaoqing(2), WOO Yat Ming Peter(3), CHEUNG Fung Ching(1)
Affiliation :
(1) Department of Neurosurgery, Queen Elizabeth Hospital, Hong Kong, (2)Department of Electrical Engineering, City University of Hong Kong, (3)Department of Neurosurgery, Kwong Wah Hospital, Hong Kong.
Introduction :
Traumatic brain injury is associated with major morbidity and mortality, with significant impact on patients as well as the healthcare system. Multiple prognostic models exist for traumatic brain injury (TBI) to predict outcome and survival, including clinical and radiological findings. However, they are not frequently applied in daily practice due to lack of prediction power and often oversimplifying the parameters, using conventional biostatistical methods and software.
Objectives :
To develop a prediction model using machine learning methods for outcome prediction in patients with moderate and severe traumatic brain injury.
Methodology :
Data was collected from the KCC TBI database consisting of data from patients suffering from moderate and severe traumatic brain injury, who were admitted to Queen Elizabeth Hospital from 2006 to 2014. Demographic data, features, and measurement on computed tomography of the brain, neurological condition on admission, admission vitals, comorbidities were analyzed. Clinical outcome was based on Glasgow Outcome Score (GOS) at 6 weeks and 6 months. Favourable outcome was defined as GOS between 4 and 5. A total of over 3400 patients were analyzed, using different methods of machine learning analysis. We present the classification performance with different methods of machine learning, including Logistic Regression (LR), Naïve Bayes (NB), Random Forest (RF), XGBoost, Support Vector Machine (SVM) and Artificial Neural Network (ANN). Seven evaluation metrics were utilized to evaluate the classification performance, including accuracy, sensitivity, specificity, precision, recall, F1-score, and the area under the curve (AUC).
Result & Outcome :
Result: After data cleansing, the data was assigned to training and testing group in approximately 4:1 ratio. For 6 weeks outcome, ANN has the highest AUC (0.91) for predicting mortality, while RF has the highest AUC (0.90) for predicting favourable clinical outcome. For 6 months outcome, XGboost has the highest AUC (0.91) for predicting mortality, while ANN has the highest AUC (0.92) for predicting favourable outcome. Conclusion: Machine learning method can predict outcome and survival of patients with traumatic brain injury, which may be more useful than traditional biostatistical methods in handling big data

Abstracts With Same Type

Abstract ID
Abstract Title
Abstract Topic
Submission Type
Primary Author
HAC6312
Clinical Safety and Quality Service II (Projects aiming to enhance clinical safety and outcomes, clinical governance / risk management)
HA Staff
Kit Ling WONG
HAC6090
Research and Innovations (new projects / technology / innovations / service models)
HA Staff
chan marko
HAC5861
Clinical Safety and Quality Service I (Projects aiming to improve efficiency and effectiveness of care delivery to meet international standards)
HA Staff
Ms. Sabrina Ho
HAC5712
HA Young Investigators Session (Projects to be presented by HA staff who had joined HA for 10 years or less)
HA Staff
Mr. CHIT YI LAU
HAC5716
Clinical Safety and Quality Service II (Projects aiming to enhance clinical safety and outcomes, clinical governance / risk management)
HA Staff
Shuk Ching MAK
HAC5675
Staff Engagement and Empowerment (motivating staff / teamwork / work revamp tackling manpower issue / staff wellness / OSH / retention)
HA Staff
Connie Suk Ling LO
HAC6327
Staff Engagement and Empowerment (motivating staff / teamwork / work revamp tackling manpower issue / staff wellness / OSH / retention)
HA Staff
Yuk Sim LUI
HAC5990
HA Young Investigators Session (Projects to be presented by HA staff who had joined HA for 10 years or less)
HA Staff
P Y SY
301 visits